You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
As a result of #666 / 7aa07ef the order of tasks in cmdDel for the bridge plugin was updated such that the bridged interface is removed before DEL is called on the ipam provider.
This causes an issue for the DHCP Ipam plugin, as it requires the link to still exist so that it can communicate with the DHCP server to release the pod's IP address.
Example logs of the failure ("no such device" is caused by a failure when calling dhcp4client.NewPacketSock on the cleaned up link):
2022/12/04 18:37:19 37060c8f178b85cdfd9a5524dbd88b630cae422c33e5c3bbb59aac6c398658ad/default/eth0: releasing lease
2022/12/04 18:37:19 37060c8f178b85cdfd9a5524dbd88b630cae422c33e5c3bbb59aac6c398658ad/default/eth0: failed to release DHCP lease: no such device
Although I understand the logic in #666, I think it would be best to revert 7aa07ef for the following reasons:
All of the other main plugins in the project (host-device, macvlan, etc) call ipam.ExecDel before their own cleanup - is there a reason the bridge plugin in particular needs different behaviour?
As pointed out in the original issue, issues cleaning up / removing the link itself are rare / impossible.
As a result of the current behaviour, the DHCP server always ends up filling up with loads of unreleased leases when pods shutdown.
cc @gojoy@squeed@mars1024@dcbw as you reviewed / discussed / authored the original issue / PR so may have better context.
The text was updated successfully, but these errors were encountered:
[This change][upstream] originally caused the bridge to clean up the
interface before calling the DHCP plugin to clean up.
Unfortuneately, this caused an error as DHCP requires the interface to
still exist so it can release the IP address to the DHCP server.
This patch reverts it as it is required in kios. Please see [the
upstream issue][issue] for further discussion on this.
[upstream]: containernetworking/plugins#702
[issue]: containernetworking/plugins#791
As a result of #666 / 7aa07ef the order of tasks in
cmdDel
for the bridge plugin was updated such that the bridged interface is removed before DEL is called on the ipam provider.This causes an issue for the DHCP Ipam plugin, as it requires the link to still exist so that it can communicate with the DHCP server to release the pod's IP address.
Example logs of the failure ("no such device" is caused by a failure when calling
dhcp4client.NewPacketSock
on the cleaned up link):Although I understand the logic in #666, I think it would be best to revert 7aa07ef for the following reasons:
ipam.ExecDel
before their own cleanup - is there a reason the bridge plugin in particular needs different behaviour?cc @gojoy @squeed @mars1024 @dcbw as you reviewed / discussed / authored the original issue / PR so may have better context.
The text was updated successfully, but these errors were encountered: